set all scores to zero for empty strings

Author: felixgwu

README.md

@@ -7,8 +7,8 @@ Automatic Evaluation Metric described in the paper [BERTScore: Evaluating Text G
   - Support 8 SimCSE models
   - Fix the support of scibert (to be compatible with transformers >= 4.0.0)
   - Add scripts for reproducing some results in our paper (See this [folder](./reproduce))
-  - Support fast tokenizers in huggingface transformers with `--use_fast_tokenizer`. Notably, you will get different scores because of the difference in the tokenizer implementations. 
-- Updated to version 0.3.9
+  - Support fast tokenizers in huggingface transformers with `--use_fast_tokenizer`. Notably, you will get different scores because of the difference in the tokenizer implementations ([#106](https://github.com/Tiiiger/bert_score/issues/106)). 
+- Updated to version 0.3.10
   - Support 3 BigBird models
   - Fix bugs for mBART and T5
   - Support 4 mT5 models as requested ([#93](https://github.com/Tiiiger/bert_score/issues/93))

bert_score/__init__.py

@@ -1,3 +1,3 @@
-__version__ = "0.3.9"
+__version__ = "0.3.10"
 from .score import *
 from .scorer import *

bert_score/utils.py

@@ -463,12 +463,14 @@ def greedy_cos_idf(ref_embedding, ref_masks, ref_idf, hyp_embedding, hyp_masks,
 
     if torch.any(hyp_zero_mask):
         print(
-            "Warning: Empty candidate sentence detected; setting precision to be 0.", file=sys.stderr,
+            "Warning: Empty candidate sentence detected; setting raw BERTscores to 0.", file=sys.stderr,
         )
         P = P.masked_fill(hyp_zero_mask, 0.0)
+        R = R.masked_fill(hyp_zero_mask, 0.0)
 
     if torch.any(ref_zero_mask):
-        print("Warning: Empty reference sentence detected; setting recall to be 0.", file=sys.stderr)
+        print("Warning: Empty reference sentence detected; setting raw BERTScores to 0.", file=sys.stderr)
+        P = P.masked_fill(ref_zero_mask, 0.0)
         R = R.masked_fill(ref_zero_mask, 0.0)
 
     F = F.masked_fill(torch.isnan(F), 0.0)

setup.py

@@ -3,7 +3,7 @@
 
 setup(
     name="bert_score",
-    version='0.3.9',
+    version='0.3.10',
     author="Tianyi Zhang*, Varsha Kishore*, Felix Wu*, Kilian Q. Weinberger, and Yoav Artzi",
     author_email="tzhang@asapp.com",
     description="PyTorch implementation of BERT score",